Climate warming is intensifying drought impacts on vegetation, especially in mountainous regions where complex topography mediates ecological responses. However, elevational variation in vegetation drought responses and their drivers remains unclear. Here, we assessed vegetation Net Primary Productivity (NPP) sensitivity to drought across elevations and vegetation types with multiple drought metrics in Nepal. Combining a structural equation model and random forest analyses, we quantified impact pathways and the relative importance of environmental drivers, providing new insights into drought impact. We found that drought frequency decreases with elevation in Nepal. Both the probability and intensity of drought-induced NPP loss showed nonlinear elevation patterns, decreasing before increasing, with inflection points around 3,200-3,600 m and 4,000-4,400 m, respectively. Although severe NPP loss was generally associated with stronger droughts, the convergence of loss thresholds at higher elevations reflects reduced ecological resilience and heightened vulnerability of alpine systems. Temperature emerged as the dominant driver of loss intensity and represented the most upstream climatic control in the pathway (mean beta = 0.37), while vegetation condition and soil moisture, respectively, dominated the indirect regulation of loss probability and intensity. However, the drivers of NPP loss reorganize along the elevation gradient, with single factor dominance at low elevations, a temperature importance peak at similar to 3,000 m, and co-regulation by multiple factors at high elevations. This study advances our understanding of nonlinear climatic stress in mountain ecosystems.